Re: [gpsd-dev] regression test puzling non-failure which should probably fail

[Eric S. Raymond]

Greg Troxel <address@hidden>:
> 
> "Eric S. Raymond" <address@hidden> writes:
> 
> >> Is it really trying the port in one half before forking?
> >
> > No, because I couldn't see any way to eliminate the TOCTOU problem.
> 
> I don't follow.   It seems that on picking a random port two potential
> realities could play out:
> 
>   The port is already in use (with odds n/16384, for n the number of
>   outgoing connections on the machine).
> 
>   The port isn't in use, and within a second becomes in use (with odds
>   n/16384 * L, where L is the poisson rate of new connections/second,
>   and these odds should be near zero).
> 
> The second is TOCTOU, and I can see that's hard to fix, but the first
> isn't a TOCTOU issue.
> 
> But I could be missing something.

You're not.  I thought the odds of a pair collision were already low
enough not to justify additional conmplexity.  But since it's come up
I might try putting retry logic in place.
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>

signature.asc
Description: Digital signature